Downloads · 30 days
333
11% of all-time downloads
EarthnDusk/MalwareDream_Final
MalwareDream_Final is a text-to-image model from EarthnDusk. Use it when you need an image from a text prompt. It is set up for diffusers. The card lists the license as creativeml-openrail-m.
Downloads · 30 days
333
11% of all-time downloads
All-time downloads
3.2K
Public
Repo size
12.5 GB
Likes
3
Public
Click a slice to open those files.
.bin5.5 GB · 45%
From the Hugging Face model README
Training settings: First run since v2: 7500 steps, then added MORE DATA of over 1000 images of various concepts and styles - trained it again at 7500 steps, but with a better learning rate at 5e6, Text Encoder steps was around 600 or more? Felt V3 wasn't stable enough, did it again for another 7500 steps, text encoder to 0 cause alread trained. All up it training time in theory was only a couple of hours, as we have access to runpod using: 1x RTX A4500, 12 vCPU 62 GB RAM, 20 GB Disk and 120 GB Network Volume. Trust me, this did actually cost us money as we didn't use Dreambooth via Colab this time, and took over 15 USD to run this, and could've maybe attemtped it with EveryDream trainer isntead, and in fact in future any updates MIGHT be done via that instead of classic dreambooth.
AI is our primary source of income. Your support is greatly appreciated!
Malware_Dream_v2 Dreambooth model trained by Duskfallcrew with TheLastBen's fast-DreamBooth notebook THE ANIME MODEL YOU NEVER ASKED FOR You are legally responsible for any wrong doings you create with this model, and we don't wanna know - so just don't.